A New Feature for Musical Genre Classification of MPEG-4 TwinVQ Audio Data

抄録

This paper proposes a new musical feature to classify MPEG-4 TwinVQ compressed data into musical genre without decoding to audio signals. To extract the musical feature, we use the LSP (Line Spectrum Pair) parameters directly extracted from a bitstream without any computation, and the Discrete Wavelet Transform (DWT). We experimented on 2,196 compressed music data collected from 10 musical genres and evaluated the performance of the musical feature for musical genre classification. The maximum of average correct ratio for musical genre classification was 81.7%. Experiment showed that the musical feature had very good performance for musical genre classification in the compressed domain of MPEG-4 TwinVQ audio compression.